1

The relationship between trade openness and economic growth: Some new insights on the

openness measurement issue

Marilyne Huchet-Bourdon, Chantal Le Mouël, Mariana Vijil

Agrocampus-Ouest, UMR 1302 SMART, Rennes, France

Inra, UMR 1302 SMART, Rennes, France

Acknowledgements : Authors thank Miao-Miao Wang for statistical assistance. They also thank Anne-Célia

Disdier, Akiko Suwa-Eisenmann, Julien Gourdon and Marcelo Olarreaga for helpful comments. This version has

benefited from the financial support of the project ANR-12-JSH1-0002-01.

Abstract

In spite of the wave of liberalization undertaken during the last decades, the debate, among

economists, on the links and causality between trade openness, growth and income distribution

is still open. Empirical results most often suggest that, in the long run, more outward-oriented

countries register better economic growth performance. However, this empirical evidence

continues to be questioned for at least two main reasons: there are still some discussions and

doubts on the way countries’ trade openness is measured on the one hand, the debate on the

estimation methodology is still open on the other hand. The aim of this paper is to contribute to

this debate by proposing a more elaborated way of measuring trade openness taking into

account two additional dimensions of countries’ integration in world trade: quality and variety.

Our results confirm that countries exporting higher quality products grow more rapidly. More

importantly, we find an interesting non-linear pattern between the trade dependency ratio and

trade in quality, suggesting that trade may impact growth negatively for countries which have

specialized in low quality products. A non-linear relationship between exports variety, trade

ratio and growth is also found, suggesting that countries exporting a wider range of products

will grow more rapidly until a certain threshold in terms of dependency of the economy to

trade.

Key words: growth, openness, quality, variety, generalized method of moments, dynamic

panel estimation

2

1 – Introduction

In spite of the wave of liberalizations undertaken during the last 30 years, the debate on the

links and causality between trade openness, growth and income distribution is still open

(Rodriguez and Rodrik, 2001). Empirical evidence tends to show that in the long run more

outward-oriented countries register higher economic growth (e.g., among others, Sachs and

Warner, 1995; Edwards, 1998; Frankel and Romer, 1999; Dollar and Kraay, 2004; Lee et al.,

2004). More recently, using broader databases and cross-section or panel-data estimations,

Freund and Bolaky (2008) and Chang et al. (2009) also show that trade openness has a positive

impact on income and that this positive relationship is enhanced by complementary policies.

According to some authors however (e.g., Rodriguez and Rodrik, 2001) most of this work

suffer from, at least, two serious shortcomings that make their results to be questioned: the way

trade openness is measured and the retained estimation methods.

Reviewing the existing literature on trade and growth shows that there is not a clear definition

of trade openness. For many authors trade openness implicitly refers to trade policy orientation

and what they are interested in is to assess the impact of trade policy or trade liberalization on

economic growth. For other authors however, trade openness is a more complex notion,

covering not only the trade policy orientation of countries but also a set of other domestic

policies (such as macroeconomic policies or institutional ones) which altogether make the

country more or less outward oriented. In such a case, what the authors are interested in is to

measure the impact of global policy orientation on economic growth. Finally, one may adopt an

even more global view of trade openness covering not only the policy dimension but also all

other non-policy factors that clearly have an impact on trade and on the outward orientation of

countries. Factors such as geography and infrastructures, for instance, do affect trade and the

outward orientation of countries, whatever their policy orientation is.

Many different measures of trade openness have been proposed and used in empirical analyses

of the relationship between openness and growth. They more or less relate to the three

alternative definitions of openness mentioned above. In line with the trade policy orientation

definition, some authors have retained measures based on trade restrictions/distortions, such as

average tariff rates1, average coverage of quantitative barriers, and frequency of non-tariff

barriers or collected tariff ratios (see, e.g., Pritchett, 1996; Harrison, 1996; Edwards, 1998,

Yanikkaya, 2003). Obviously, these indicators are very imperfect and partial measures of the

1 And/or other characteristics of the tariff distribution: tariff dispersion, frequency of tariff picks, etc.

3

overall restrictions/distortions induced by trade policies. Furthermore, data required to compute

such indicators are often available for only a limited set of countries and years.

In terms of the global policy orientation definition, various “qualitative” indices allowing for

classifying countries according to their trade and global policy regime have been proposed (see,

e.g., the 1987 World Development Report outward orientation index or the openness indices

proposed by both Sachs and Warner, 1995, and Wacziarg and Welch, 2003). Such measures

unfortunately provide only a very rough classification of countries (from rather closed to rather

open). Also many of the data required to construct these indices are available only for a few

countries and at one point in time.

Finally, measures based on trade flows, which have been commonly used in empirical analyses,

rather relate to the most global definition of trade openness. Trade dependency ratios are the

most popular of these measures (see, e.g., Frankel and Romer, 1999; Irwin and Tervio, 2002;

Frankel and Rose, 2002; Dollar and Kraay, 2004 and Squalli and Wilson, 2011, for a recent

contribution). Their main advantage is that the data required to compute them are available for

nearly all countries and over a rather long period. Their main weakness is that they are mainly

outcome-based measures, and as such, are the result of very complex interactions between

numerous factors so that it is not clear what such measures exactly capture. Another limitation

of these trade dependency ratios lies in their endogeneity in growth regressions, which requires

specific estimation techniques (such as instrumental variables techniques as in Frankel and

Romer, 1999, and Irwin and Tervio, 2002, or identification through heteroskedasticity

techniques as in Lee et al., 2004).

This last limitation may in fact be extended to all trade openness measures, and constitutes the

second shortcoming in existing empirical evidence that has been pointed out by Rodriguez and

Rodrik (2001). As argued by Lee et al. (2004), all measures of openness are generally closely

linked to the growth rate. Hence, it is likely that all measures of openness are jointly

endogenous with economic growth, which may cause biases in estimation resulting from

simultaneous or reverse causation. Various methods have been used to remedy this problem

and there is still a debate among scientists about which method is the most appropriate (see,

e.g., Dollar and Kraay, 2004; and Lee et al., 2004).

In this paper, our aim is to contribute to the on-going debate on the growth effect of trade by

enriching the most global definition of trade openness. We argue that trade openness is a

multidimensional concept that cannot be summarized to a single measure such as the

commonly used trade ratio. Thus, following recent developments in growth theory and in

4

international economics, we propose a more elaborated way of measuring trade openness

taking into account two additional dimensions of countries’ integration in world trade: the

quality and the variety of the exported basket. Indeed, according to the existing literature both

these factors are likely to affect positively growth, which call for considering them when

measuring countries’ trade openness in view of examining the relationship between trade and

growth.

On the one hand, endogenous growth theory has provided a framework for a positive growth

effect of trade through innovation incentives, technology diffusion and knowledge

dissemination (see, e.g., Young, 1991; Grossman and Helpman, 1991). Inspired from these

theoretical developments, Hausmann et al. (2007) proposed an analytical framework linking

the type of goods (as defined in terms of productivity level) a country specializes in to its rate

of economic growth. In order to test empirically for this relationship, they defined an index

aiming at capturing the productivity level (or the quality) of the basket of goods exported by

each country. Using various panel data estimators during the period 1962 – 2000, their growth

regression showed that countries exporting goods with higher productivity levels (or higher

quality goods) have higher growth performances. These results suggest that what countries

export matters as regards the growth effect of trade. Hence, our measurement of trade openness

should consider this quality dimension as a complement to the trade ratio (or the dependency)

dimension.

On the other hand, monopolistic competition trade models with heterogeneous firms and

endogenous productivity provide theoretical support for a positive impact of trade openness on

growth. Indeed, the theory predicts a productivity improvement in the country due to the exit

of less efficient firms after trade liberalization -or a reduction in transport costs for example-

(e.g., Melitz, 2003). Furthermore, a higher share of the most productive firms will start

exporting, which translates into an increase in the variety of exports. As exporters are more

productive on average than domestic firms, an increase in exports variety can be associated to

rising country productivity.

Based on this literature, Feenstra and Kee (2008) developed a model allowing to link, across

countries and over time, relative export variety to total factor productivity using a GDP

function. They tested this relationship on the basis of exports to the US for a panel of 48

countries over the period 1980-2000 using three stage least squares regressions. Their empirical

results indicated that there is a positive and significant relationship between export variety and

average productivity. Furthermore, computing the gains from trade in the monopolistic

5

competition model of Melitz (2003), Feenstra (2010) shows that countries with a greater export

over GDP ratio will experience higher gains in terms of GDP per capita growth, from export

variety. Once again, these results suggest that, in addition to the trade dependency ratio, the

structure of countries’ exports matters regarding the growth effect. Hence, our measurement of

trade openness should also consider this variety dimension.

Our empirical application draws on the Barro and Lee (1994)’s model, which has been

extended to take into account our set of three indicators of trade openness: trade dependency

ratio, quality index and variety index. Barro and Lee (1994) study empirical determinants of

growth. They are in line with the endogenous growth theory. Unlike the usual neoclassical

growth model for a closed economy (Solow, 1956), endogenous growth models take into

account the sources of technological progress (human capital, role of government for instance).

Thus, we include some proxies for trade openness in our empirical model as potential sources

of technological change.

Estimations are performed on 5-year averaged data over the period 1980-2004 for an

unbalanced panel of 158 countries. We use a Generalized Method of Moments (GMM)

estimation approach developed for dynamic panel data models in order to deal with the

potential endogeneity bias due to omitted variables, simultaneity and measurement error.

Our results confirm that countries more open to trade and exporting higher quality products

experience higher growth. More importantly, we point out an interesting pattern of non

linearity in the growth effect of the trade ratio: the higher the quality of the export basket of the

country, the greater the positive impact of trade on economic growth. In addition, there is a

minimum level of export quality under which trade can be detrimental to growth. This non-

linear pattern in the trade to growth relationship is found for the whole sample and for various

sub-samples of developing countries. It has particularly important implications for developing

countries since as they often exhibit low quality export baskets, they are more likely to

experience negative trade impact on growth.

From our estimation results we also confirm a non linear relationship between the export

variety, the trade ratio and growth. Export variety has often a positive impact on growth per se;

but this relationship seems to exist until a certain degree of dependency of the economy on

trade. As most developing countries are below this threshold, export diversification appears as

an important strategy for them.

6

The remainder of the article is organized as follows. In the next section, we present the

specification of performed growth regressions and the retained econometric methodology.

Section 3 reports and discusses empirical results, while section 4 concludes.

2 – Specification of growth regressions and econometric methodology

Inspired from Barro and Lee (1994)’s approach we retain the following specification:

titititi

titi

titiGDP

X

GDP

Ilifeeducation

pop

GDP

pop

GDP,

,4

,31,21,1

1,,

)ln(lnln νγµββββα +++

+

+++

=

−−

−

(1)

where the dependent variable is the logarithm of GDP per capita of country i for period t, with

GDP corresponding to Gross Domestic Product and pop to the total population. Explanatory

variables are the following. First, the initial level of GDP per capita is included to test for the

impact of initial conditions. Countries’ endowments in production factors are controlled for

using the initial level of human capital investment, which is approximated through the level of

education (education) and the life expectancy at birth (life); and the physical investment as

measured by the investment over GDP ratio

GDP

I 2. The effects of education, life

expectancy and investment ratio are likely to be positive. Finally, in order to test for the impact

of trade on income per capita, we choose as a measure of trade openness the export ratio

(GDP

X, i.e., exports over GDP), an export quality index )(Quality , an export variety index

)(Variety and the combined effect of the export ratio with each of these indices. We decided to

choose the export ratio instead of the usual trade ratio (GDP

MX +, i.e, sum of exports and imports

over GDP) in order to keep consistency with the quality and the variety indices which are

concerned with growth mechanisms arising from the export side.

Thus, our two alternative specifications are:

2 Due to the lack of available data, general government final consumption expenditure ratio, black market premium and revolution variables used by Barro and Lee (1994) are not introduced here.

7

- an extended specification including the export quality index (Quality) and its cross impact

with the export ratio:

titititi

ti

titititi

titi

QualityGDP

XQuality

GDP

X

GDP

Ilifeeducation

pop

GDP

pop

GDP

,,,

6,5

,4

,31,21,1

1,,

)ln(*)ln(

)ln(lnln

νγµββ

ββββα

+++

++

+

+++

=

−−

− (2)

- an extended specification with the alternative export variety index (Variety) and its cross

impact with the export ratio:

titititi

ti

titititi

titi

VarietyGDP

XVariety

GDP

X

GDP

ILifeeducation

pop

GDP

pop

GDP

,,,

6,5

,4

,31,21,1

1,,

)(*)(

ln)ln(lnln

νγµββ

ββββα

+++

++

+

+++

=

−−

− (3)

The model includes time-specific effects ( tγ ) accounting for period-specific effects such as

productivity changes that are common to all countries or the global effect of US dollar

appreciation, country-specific fixed-effects ( iµ ) that take into account country-specific

features that are constant in time, such as geography, and an error term ( ti ,ν ).

Our empirical estimation is run on an unbalanced panel of 158 countries for the period 1980-

2004 using 5-year averaged data (except for initial GDP per capita, education and life

expectancy that take the first observation within each period). As most explanatory variables

are likely to be jointly endogenous with economic growth while important variables, e.g., the

country-specific effects, are not observable and omitted in the estimation, estimating this model

by Ordinary Least Squares (OLS) or Within group estimations would potentially lead to biased

results. Thus, we use the System-GMM estimator developed for dynamic panel data models

(Arellano and Bover, 1995; Blundell and Bond, 1998). The main advantage of this estimator is

that it does not require any external instrument to deal with endogeneity.

Within the GMM approach, one may choose the first-differenced estimator, which considers

regression equations in first-differences instrumented by lagged levels of explanatory variables.

Taking first-differences eliminates country-specific fixed-effects, thus solving the problem of

the potential omission of time invariant country specific factors that may influence growth.

Nevertheless, the first-differenced GMM estimator (Arellano and Bond, 1991) is not suitable

when time series are persistent and the number of time series observations is small, like in the

8

case of empirical growth models where data has to be averaged3 in order to avoid modelling

cyclical dynamics (Bond et al., 2001). Under these conditions, lagged levels of explanatory

variables tend to be weak instruments for subsequent first-differences, thus producing biased

estimates. Therefore, Arellano and Bover (1995) and Blundell and Bond (1998) suggest to

retain the System-GMM approach, which combines - into one system - regression equations in

first-differences and in levels, where instruments used for level equations are lagged first-

differences of the series.

Hence, departing from this general model:

tiittititi Xyy ,,1,, ' νητβα ++++= − for i =,…, N and t =2,…, T (4)

where

tiiti ,, νηε += has the standard error component structure:

[ ] [ ] [ ] 0. ,, =Ε=Ε=Ε tiitii νηνη for i =,…, N and t =2,…, T (5)

y is the dependent variable, X is the vector of explanatory variables, iη and tτ denote

respectively unobserved country- and time-effects and ti ,ν is the idiosyncratic disturbance term.

We perform the following transformation to remove the unobserved individual effect:

)()()(')( 1,,11,,2,1,1,, −−−−−− −+−+−+−=− tititttitititititi XXyyyy ννττβα (6)

Nevertheless, instead of using a “first-difference transformation” as is usually done, we

perform a “forward orthogonal deviation”. Thus, instead of subtracting the previous

observation from the contemporaneous one, we subtract the average of all future available

observations of a variable4. This way of dealing with heterogeneity allows us to preserve

sample size in our unbalanced panel while still being able to use past values of explanatory

variables as instruments (Arellano and Bover, 1995; Roodman, 2006).

3 Data is usually averaged over 5 years.

4 That is, for a variable w the transformation will be:

−≡ ∑

⟩+

tsis

itititti w

Twcw

11, where the sum is taken

over all available future observations itT , and the scale factor itc is )1( +itit TT .

9

Under the assumption of absence of serial correlation in the idiosyncratic disturbance terms on

the one hand:

[ ] 0. ,, =Ε siti νν for i = 1, …., N and ts ≠ , (7)

that the initial conditions are predetermined on the other hand:

[ ] 0. ,1, =Ε tiiy ν for i =,…,N and t =2,…, T, (8)

the differenced equation (6) can be instrumented by lagged levels of explanatory variables

(Arellano and Bond, 1991), using the following � = 0.5�� − 1�� − 2moment conditions:

( )[ ] 0. 1,,, =−Ε −− titistiy νν (9)

( )[ ] 0. 1,,, =−Ε −− titistiX νν (10)

For t =3,…, T and 2≥s

Furthermore, according to Blundell and Bond (1998), when combining equations (4) to (8)

with two additional assumptions:

� iη . ( )1,2, ii yy − � = 0 (11)

� iη . ( )1,2, ii XX − � = 0 for i =,…, N (12)

which are restrictions on the initial conditions of the data generating process5; � − 2 additional

moment conditions can be used:

( )[ ] 0. 2,1,, =−Ε −− tititi yyε (13)

( )[ ] 0. 2,1,, =−Ε −− tititi XXε for i =,…, N and t =3,…, T (14)

This allows the use of lagged first-differences of the series as instruments for equation in

levels, as suggested by Arellano and Bover (1995).

5 In our context, assumption (11) means for example that deviations of 1,iy from long-run steady-state values must

not depend on unobserved fixed-effects, even if the latest can affect the level of steady-state outputs. Bond et al.

(2001) argue that this assumption may be valid in growth model frameworks, thus allowing us to use the System-

GMM estimator in our model.

10

Thus, System-GMM estimator implies running a GMM procedure on the following system of equations:

)()()(')( 1,,11,,2,1,1,, −−−−−− −+−+−+−=− tititttitititititi XXyyyy ννττβα (7)

and

tiittititi Xyy ,,1,, ' νητβα ++++= − (16)

In order to test for the appropriateness of our retained instruments, we consider two

specification tests. The first one is the Hansen test of over-identification for which the null

hypothesis is that the chosen instruments are valid. The second one examines whether the

idiosyncratic disturbance term ti ,ν is serially correlated. The test is performed on the first-

differenced error term (that is, the residual of equation (7)) and the null hypothesis is that the

latter is second-order uncorrelated. In both cases, failure to reject the null hypothesis gives

support to our retained specification.

3 – Data and results

3.1. Data

To reduce the impact of business cycles, we use a total of five-year averaged data between

1980 and 2004 for an unbalanced panel of 158 countries (Appendix A provides the full list of

countries in the sample). Most required data are extracted from the World Bank World

Development Indicators (WDI) database, as it is the case for the following variables. The

dependent variable is computed using the GDP per capita based on purchasing power parity

(expressed in constant 2005 US dollars). The investment ratio is proxied through the gross

fixed capital formation in percentage of GDP; the life expectancy at birth is the number of

years one is expected to stay alive when birthing; and the education level is measured as the

gross secondary school enrolment ratio. The export ratio is computed using GDP as well as

values of exports of goods and services in current US dollars.

The export quality index is computed according to the Haussmann et al. (2007)’s approach and

the variety indicator according to Feenstra and Kee (2008) and Feenstra (2010). They are both

computed based on export values in current US dollars extracted from the CEPII international

trade database BACI (at a SITC2 disaggregated level). Further details on the definition and

computation of these indicators can be found in appendix B. Table 1 summarizes some basic

descriptive statistics for all variables used in growth regressions.

11

Table 1: Descriptive statistics for variables used in the model

Variables Obs Mean Std. Dev. Min Max

GDP per capita (constant 2005 USD) 756 9308.66 11578.08 108.21 94734.24

Education (%) 896 60.14 32.74 2.40 161.74

Life expectancy (years) 1147 64.62 10.21 30.47 81.08

Investment / GDP (%) 709 21.62 7.84 2.53 86.79

Exports / GDP (%) 736 35.22 24.03 2.76 199.12

Export quality (current USD) 756 7808.01 3681.57 1771.54 27594

Export variety (%) 756 66.94 27.02 5.11 1

Source: Authors’ calculations

3.2. Empirical results

In this section we examine whether trade openness can be considered as a main determinant of

growth. Results of the first regression (1) are reported as a benchmark in Table 2 since they

refer to the specification including the export ratio as the single measure of trade openness. The

quality and the variety index as additional measures of trade openness are presented in columns

(2) and (4) respectively. Results of the regressions including in addition the cross effect of the

export ratio with the quality and with the variety index are reported in columns (3) and (5)

respectively.

As far as the first specification is concerned, Table 2 indicates that when trade openness is

measured by the export ratio only, it does not appear as a significant determinant of growth. In

line with Rodriguez and Rodrick (2001), this could be caused by two technical issues ; the first

one being the endogeneity of trade as regard to growth and the second one being the way

openness to trade is measured. We are confident that our empirical strategy allows us to deal

properly with any kind of endogeneity. What we are interested in is then to check if this lack of

statistical impact is originating from an ill-specified measure of trade integration.

Interestingly, column (2) shows that when trade openness is measured by both the export ratio

and the export quality index, the latter only has a positive and significant impact on GDP per

12

capita growth. This result confirms Hausmann et al. (2007)’s result that a higher quality of

exports enhances growth.

Finally, column (3) reveals an interesting non-linear pattern between trade openness and

growth once the export ratio is crossed with the quality index, as both this variable and the

export ratio appear statistically significant. Our estimation results suggest that trade may have a

negative impact on growth when countries have specialized in low quality products; trade

clearly enhances growth once countries have specialized in high quality products and their

export basket exhibits a minimum required level of quality. The corollary is also true, as the

higher the quality of the export basket, the greater the impact of the export ratio on growth.

More specifically, Table 2 indicates that, all other things being unchanged, one percentage

point increase in the export ratio would raise the 5-years average GDP per capita by (-0.057 +

0.006*LnQuality). Hence, a minimum level of quality of the export basket is required (13 360

current USD) for the impact of the export ratio starts to be positive. As indicated by Table 1,

this threshold is much higher than the average of the export quality index over the whole

sample (7 808 current USD) suggesting that trade is likely to enhance growth only for countries

which are used to exhibit high quality of export baskets.

Once we exclude major oil exporting countries from the sample (column (3_o)), 6 results

remain similar to those of specification (3), suggesting no specific behaviour for these

countries. It is interesting to note that the minimum level of quality of the export basket

required for a positive impact of the export ratio on the GDP per capita growth remains

unchanged at 13 360 current USD.

Turning now to the variety dimension, specification (4)’s results show that when trade

openness is measured by both the export ratio and the export variety index, the latter only has a

positive and significant impact on GDP per capita. This result is in line with Feenstra and Kee

(2008) and Feenstra (2010) which suggest that a higher variety of exports contributes to

enhance growth.

However, the cross effect of the export ratio and the variety index does not appear statistically

significant (column 5), while the impact of export variety remains so. Hence our results seem

6 A country is considered to be a major oil exporting country if on average, over the 1980-2004 period, the value of its oil exports account for more than 2/3 of the value of its total exports ( these countries represents 10% of the whole sample). One must underline that results are robust to changing this oil over total exports threshold to 50%.

13

to suggest that when the variety of exports is considered, there are no complementarities

between this and the export ratio; only the variety has a positive impact on growth.

Nevertheless, results could be biased by the presence in our sample of oil exporting countries

which exhibit particularly low export variety indices and high export ratios.

Indeed, when major oil exporting countries are excluded from the sample (specification (5_o)),

one recovers the non-linear impact of trade on growth: the cross effect of the export ratio and

the variety index becomes negative and statistically significant, and both the variety index and

the trade ratio appear positive and significant. Results indicates that, all other things being

unchanged, one percentage point increase in the export ratio would raise the 5-years average

GDP per capita by (0.010 - 0.011*Variety). Hence, a maximum level of variety of the export

basket is required (0.90) for the impact of the export ratio to remain positive. Since, as

indicated by Table 1, most observations of our sample are below this threshold, we can

conclude that the export ratio has nearly always a positive effect on GDP per capita. The

corollary would be that the impact of an increase in the export variety on growth is positive

until a certain degree of dependency of the economy on exports (the export ratio has to remain

below 51%). As indicated by Table 1, this threshold is higher than the average of the export

ratio over the whole sample (35.22%).

Regarding control variables, Table 2 shows that initial GDP per capita exhibit an expected

positive and close to 1 statistically significant coefficient. Among the main growth

determinants considered by Barro and Lee (1994), the investment ratio has an expected positive

and significant impact in most of the specifications. In terms of human capital, nor the

secondary enrolment ratio nor the life expectancy at birth have a significant impact on GDP per

capita growth. This is puzzling but may be due to the fact that these variables have a long term

impact on development and not a contemporaneous one. Finally, it should be noted that for all

estimations, Hansen and AR(2) specification tests give support to our retained GMM-System

estimator. The lagged variables that are chosen appear as good instruments in the present

context.

Table 2: Growth regressions results using System-GMM estimator Total sample Without oil Ln (GDP/pop) final (1) (2) (3) (4) (5) (3_o) (5_o) Ln (GDP/pop) init. 1.089*** 0.983*** 0.924*** 1.022*** 1.019*** 0.930*** 0.993***

14

(0.076) (0.054) (0.041) (0.064) (0.042) (0.043) (0.031)

Education -0.002 -0.001 -9.35e-05 -0.007 -0.000 0.000 -0.000

(0.002) (0.002) (0.001) (0.002) (0.001) (0.001) (0.001)

I/GDP 0.026** 0.010 0.011 0.025** 0.022*** 0.0070** 0.002

(0.012) (0.007) (0.007) (0.01) (0.007) (0.003) (0.004)

Ln (life expec.) -0.809 -0.512 -0.160 -0.622 -0.678* -0.174 -0.297

(0.603) (0.456) (0.184) (0.43) (0.347) (0.194) (0.199)

X/GDP 0.002 0.001 -0.057* 0.000 0.006 -0.038* 0.010**

(0.003) (0.001) (0.032) (0.001) (0.005) (0.022) (0.004)

Ln (quality)

0.297** 0.117

0.130

(0.124) (0.107)

(0.112)

X/GDP* Ln (quality)

0.006*

0.004*

(0.003)

(0.002)

Variety

0.299* 0.566*

0.562**

(0.181) (0.310)

(0.242)

X/GDP* variety

-0.008

-0.011**

(0.006)

(0.004)

Constant 2.004 -0.481 0.105

1.778* 0.039 0.819

(1.764) (1.287) (0.829)

(1.069) (0.964) (0.688)

Observations 636 636 636 636 636 575 575

Number of panelid 158 158 158 158 158 140 140

AR(2) test p-value 0.144 0.224 0.204 0.12 0.134 0.439 0.532 Hansen test p-value 0.583 0.199 0.172 0.25 0.373 0.229 0.106 Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Estimation method: two-step GMM system (Arellando and Bover, 1995; Blundell and Bond, 1998) with Windmeijer (2005) correction and orthogonal deviation. Weakly exogenous variables used as instruments are education and life expectancy 2nd lag (3rd lag for column (3_o)). and investment, export ratio, export quality and multiplicative interaction terms 3rd lag. Exogenous variables used as instruments are year dummies (Roodman, 2006) for the system; and the predetermined variable initial GDP per capita which is only used for the level equation. Source: Authors’ calculations. To check the robustness of our results we performed the same regressions on various sub-

samples covering different groups of developing countries defined according to the 2005 World

Bank classification (Table 3). We work first with the sub-sample of Developing Countries

(DC)7; and within it, with Low Income Countries (LIC) and Lower Middle Income Countries

(LMIC) 8. As done previously, we also consider the corresponding sub-samples excluding the

major oil exporting countries.

For the estimations with the quality index crossed with the export ratio, results obtained for

these sub-samples are similar to previous one. In particular, we recover the non-linear pattern

7 Countries with a real GDP per capita in 2005 below 10 065 USD.

8 Countries whit a real GDP per capita in 2005 below 3 255 USD.

15

between trade openness and growth. Estimates indicates that the minimum level of quality of

the export basket required for the impact of the export ratio starts to be positive are 4 649 USD,

4 914 USD and 3 966 USD for respectively DC without major oil exporting countries,

LIC&MLIC and LIC&MLIC without major oil exporting countries. Taking the last period of

our sample (2000-2005), countries below this threshold are mainly African least developed

countries9. This suggest that increasing the dependency of their economy to trade without

ensuring an improvement of the quality of their exports may have negative consequences in

terms of growth. Thus, a strategy to add value-added to trade seems crucial for them.

As for estimations with the variety index crossed with the export ratio, results are robust for

LIC&MLIC; estimates confirm the non-linear relationship between trade dependency, export

variety and growth. Indeed, specifications (5’_o), (5’’) and (5’’_o) show that while the export

variety index has no longer a significant impact on GDP per capita alone, the cross effect with

the export ratio is now positive and significant. These results suggest that for the LIC&LMIC

group, trade dependency and variety contribute jointly to increase GDP per capita.

9 Burundi, Benin, Burkina Faso, Central African Republic, Comoros, Ethiopia, Guinea, Mali, Malawi, Solomon

Islands, Chad, Uganda, and Congo, Dem. Rep.

16

Table 3: Robustness analysis using various sub-samples of developing countries

DC LIC&LMIC DC without oil LIC&LMIC without oil Ln (GDP per capita final) (3’) (5’) (3’’) (5’’) (3’_o) (5’_o) (3’’_o) (5’’_o)

Ln (GDP per capita init.) 1.297*** 1.227*** 0.954*** 0.988*** 0.944*** 1.026*** 0.922*** 0.998***

(0.409) (0.151) (0.060) (0.074) (0.087) (0.093) (0.056) (0.096) Education -0.008 -0.004 -0.005** -0.002 -0.004** -0.003 -0.006*** -0.005*

(0.016) (0.003) (0.002) (0.002) (0.002) (0.003) (0.001) (0.003) I/GDP 0.023 0.008 0.012*** 0.013*** 0.014*** 0.013*** 0.012*** 0.013***

(0.014) (0.014) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Ln (life expec.) -1.858 -3.205** 0.557** -0.227 0.331 -0.246 0.591** -0.266

(1.261) (1.253) (0.245) (0.590) (0.373) (0.686) (0.237) (0.800) X/GDP 0.025 0.016 -0.051** -0.002 -0.076*** -0.000 -0.058*** 0.001

(0.059) (0.010) (0.024) (0.003) (0.021) (0.002) (0.020) (0.003) Ln (quality) 0.263 -0.096 -0.137 -0.135

(0.432) (0.083) (0.088) (0.102) X/GDP* Ln (quality) -0.002 0.006** 0.009*** 0.007***

(0.006) (0.002) (0.002) (0.002) Variety 2.091** 0.031 0.127 0.235

(0.881) (0.176) (0.225) (0.236) X/GDP* variety -0.007 0.008** 0.005 0.008*

(0.014) (0.004) (0.005) (0.004) Constant 2.651 9.859** -1.166 0.746 0.116 0.529 -0.817 0.742

(2.762) (4.003) (1.032) (1.813) (1.296) (2.153) (1.031) (2.485)

Observations 462 462 265 265 420 420 239 239 Number of panelid 120 120 74 74 107 107 65 65 AR(2)test p-value 0.108 0.289 0.128 0.099 0.353 0.253 0.332 0.291 Hansen test p-value 0.158 0.779 0.108 0.146 0.053 0.001 0.167 0.116 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Estimation method: two-step GMM system (Arellando and Bover, 1995; Blundell and Bond, 1998) with Windmeijer (2005) correction and orthogonal

17

deviation. All instruments are collapsed (Roodman, 2006). All estimations are run using the 1rts lag and further of weakly exogenous variables (as defined in Table 2)10. Exogenous variables used as instruments are year dummies (Roodman, 2006) for the system and the predetermined variable initial GDP per capita for the level equation. Source: Authors’ calculations.

10 Except for specifications (3’)-(5’) that use only the 3rd lag for all instruments and (5’’_o) which uses only the 1rst lag for the investment ratio, export ratio, export

variety/quality and the multiplicative interaction term as instruments.

18

4 - Conclusion

This paper investigates the relationship between trade openness and growth. Starting from the

idea that trade openness cannot be fully characterized through trade flows only we propose to

account for two additional dimensions of countries’ trade integration: export quality and export

variety. Then, following Barro and Lee (1994), standard growth regressions are performed

where, among the explanatory variables, the commonly used trade ratio (here the export ratio)

is complemented by the Haussmann et al. (2007)’s export quality index or the Feenstra and

Kee (2008)’s export variety index. Our empirical application is based on annual data over the

period 1980-2004 for an unbalanced panel of 158 countries. As most explanatory variables are

likely to be jointly endogenous with economic growth, we use the system GMM estimator

developed for dynamic panel data models.

Our empirical results are in line with New International Economics insights that regarding the

relationship between trade openness and growth in addition to the trade ratio, the quality and

the variety of the export basket matter. We point out an interesting non-linear pattern between

trade openness and growth when export quality is taken into account: trade may have a

negative impact on growth when countries have specialized in low quality products; trade

clearly enhances growth once countries have specialized in high quality products and their

export basket exhibits a minimum required level of quality. Therefore, there is some pattern of

complementarity between trade dependency and trade in quality so that the higher the quality

of the export basket, the greater the impact of the export ratio on growth.

Estimation results also suggest a non-linear relationship between trade and growth when the

variety of exports is taken into account. However, the impact of an increase in the export

variety on growth seems positive until a certain degree of dependency of the economy on

exports. For most developing countries, we find some pattern of complementarity between

trade dependency and variety: the export ratio has a positive impact on GDP per capita and the

higher the variety of the export basket, the higher the impact of the trade ratio. It is interesting

to note that the cross effect of the trade ratio and trade in variety clearly relates to changes at

the intensive margin and at the extensive margin of trade (even if our export ratio cannot be

properly disentangled between the two margins). Hence, further investigations are required to

clarify the role of trade dependency and trade in variety as regards the relationship between

trade and growth.

19

From an economic policy perspective, these results are very interesting as they show that

investment in productive capacity to move developing countries ‘exports up the quality chain

could be decisive to enhance growth. Also, they suggest that facilitating access to the export

market for new exporters, through export promotion agencies for example, can have important

implications for development. As aid for trade, and in particular aid for building productive

capacity, intends to focus on these matters, further evidence on its link with the quality and the

variety of exports seems to be necessary.

20

References

Arellano, M. and Bond, S. (1991), “Some tests of specification for panel data: Montecarlo

evidence and an application to employment equations”. Review of Economic Studies,

58(2), p. 277–297.

Arellano, M. and Bover, O. (1995), “Another look at the instrumental-variable estimation of

error-components models”. Journal of Econometrics, 68(1), p. 29–52.Blundell, R. and

Bond, S. (1998), “Initial conditions and moment restrictions in dynamic panel data

models”. Journal of Econometrics, 87(1), p. 115–143.

Barro R. J. and Lee J-W. (1994), “Sources of economic growth”, Carnegie-Rochester

Conference Series on Public Policy, 40, p.1-46.

Blundell, R. and Bond, S. (1998), GMM estimation with persistent panel data: an application to

production functions, The Institute for Fiscal Studies (IFS) Working Paper Series No.

W99/4.

Bond, R., Hoeffler, A. and Temple, J., (2001) “GMM estimation of empirical growth models”,

CEPR Discussion Paper Series, No. 3048.

Chang R., Kaltani L. and Loayza, N.V. (2009), “Openness can be good for growth: the role of

policy complementarities”. Journal of Development Economics, 90, p. 33-49.

De Melo, J. and Grether, J. M. (2000), Commerce International : Théories et Applications. De

Boeck Université.

Dollar, D. and Kraay, A. (2004), “Trade, Growth, and Poverty”. The Economic Journal, 114,

February, p. 22-49.

Edwards, S. (1998), “Openness, productivity and growth: what do we really know?”. The

Economic Journal, 108, March, p. 383-98.

Feenstra, R. C. (1994), “New product varieties and the measurement of international prices”.

American Economic Review, 84(1), 157–177.

Feenstra, R. C and Kee, H. L. (2008), “Export variety and country productivity: Estimating the

monopolistic competition model with endogenous productivity”. Journal of International

Economics, 74, p. 500–518.

Feenstra, R. C. (2010), “Measuring the gains from trade under monopolistic competition”,

Canadian Journal of Economics, 43 (1), p. 1-28.

Frankel, J. and Romer, D. (1999), “Does Trade Cause Growth?”. American Economic Review, 89(3), p. 379-99.

Frankel, J. and Rose, A. (2002), “An Estimate of the Effect of Common Currencies on Trade

and Income”. The Quarterly Journal of Economics, 117(2), p. 437-66.

21

Freund, C. and Bolaky, B. (2008), “Trade, regulations, and income”. Journal of development

economics, 87, p. 309-321.

Grossman, G. and Helpman, E. (1991), Innovation and growth in the World economy.

Cambridge, MA: MIT.

Harrison, A. (1996), “Openness and growth: A time-series, cross-country analysis for

developing countries. Journal of Development Economics, 48, p. 419-447.

Haussmann, R., Hwang, J. and Rodrik, D. (2007), “What you export matters”. Journal of

Economic Growth, 12, p. 1-25.

Hummels, D. and Lenow, P.J. (2005), “The variety and quality of a nation’s exports”, The

American Economic Review, 95 (3), p. 704-723.

Irwin, D. and Tervio, M. (2002), “Does Trade Raise Income? Evidence from the Twentieth

Century”. Journal of International Economics, 58(1), p. 1-18.

Lee, H.Y., Ricci, L.A., and Rigobon, R. (2004), “Once Again, Is Openness Good for Growth?”.

NBER Working Papers. Cambridge: NBER.

Lederman, D. and Maloney, W. F. (2008), “In search of the missing resource curse”. Policy

Research paper 4766, World Bank, Washington.

Melitz, J.M. (2003), “ The impact of trade on intra-industry reallocations and aggregate

industry productivity”. Econometrica, 71(6), p. 1695–1725.

Pritchett, L. (1996), “Measuring Outward Orientation in LDCs: Can It Be Done?”. Journal of

Development Economics, 49, p. 307-35.

Rodriguez, F. and Rodrik, D. (2001), “Trade Policy and Economic Growth: A Skeptic's Guide

to the Cross-National Evidence”. In NBER (ed.), NBER Macroeconomics Annual 2000,

NBER Chapters, (Cambridge: National Bureau of Economic Research), 261-338.

Roodman, D. (2007), “A Note on the Theme of Too Many Instruments”, Center for

Global Development Working Paper, No. 125

Sachs, J. D. and Warner, A. M. (1995), “Natural Resource Abundance and Economic

Growth”. NBER Working Papers, No. 5398 (New York: National Bureau of Economic

Research).

Solow, R.M. (1956), “A contribution to the theory of economic growth”, Quarterly Journal of

Economics, 70, p. 65-94.

Squalli J. And Wilson K. (2011), “A new measure of trade openness”, The World Economy,

34(10), 1745-1770.

Wacziarg, R., and Welch, K. H. (2003), “Trade Liberalization and Growth: New Evidence”, NBER

Working Paper 10152 (Cambridge: National Bureau of Economic Research).

22

Windmeijer, F. (2005). A finite sample correction for the variance of linear efficient two-step GMM

estimators. Journal of Econometrics, 126, p. 25-51.

World Development Report (1987), Oxford University Press, 301 pages.

Yanikkaya, H. (2003), “Trade openness and economic growth: A cross-country empirical

investigation”. Journal of Development Economics, 72, p. 57-89.

Young, A. (1991), “Learning by doing and the dynamic effects of international trade”,

Quarterly Journal of Economics, 106, p. 369–405.

23

Annex A: list of countries Afghanistan Angola Albania United Arab Emirates Argentina Armenia Antigua and Barbuda Australia Austria Azerbaijan Burundi Benin Burkina Faso Bangladesh Bulgaria Bahrain Bahamas, The Belarus Belize Bolivia Brazil Barbados Brunei Darussalam Bhutan Central African Republic Canada Switzerland Chile China Cote d'Ivoire Cameroon Congo, Rep. Colombia Comoros Cape Verde Costa Rica Cyprus Czech Republic Germany Djibouti Dominica Denmark Dominican Republic Algeria Ecuador

Egypt, Arab Rep. Eritrea Spain Estonia Ethiopia Finland Fiji France Gabon United Kingdom Georgia Ghana Guinea Gambia, The Guinea-Bissau Equatorial Guinea Greece Grenada Guatemala Guyana Hong Kong SAR, China Honduras Croatia Hungary Indonesia India Ireland Iran, Islamic Rep. Iceland Israel Italy Jamaica Jordan Japan Kazakhstan Kenya Kyrgyz Republic Cambodia Kiribati Korea, Rep. Lao PDR Lebanon Liberia Libya St. Lucia

Sri Lanka Lithuania Latvia Morocco Moldova Madagascar Maldives Mexico Macedonia, FYR Mali Mongolia Mozambique Mauritania Mauritius Malawi Malaysia Niger Nicaragua Netherlands Norway Nepal New Zealand Oman Pakistan Panama Peru Philippines Papua New Guinea Poland Portugal Paraguay Qatar Romania Russian Federation Rwanda Sudan Senegal Solomon Islands Sierra Leone El Salvador Suriname Slovak Republic Slovenia Sweden

Seychelles Syrian Arab Republic Chad Togo Thailand Tajikistan

Tonga Trinidad and Tobago Tunisia Turkey Tanzania Uganda

Ukraine Uruguay United States Uzbekistan St. Vincent and the Grenadines Venezuela, RB

24

Vietnam Vanuatu

Yemen, Rep. South Africa

Congo, Dem. Rep. Zambia

25

Annex B: The quality and variety indices The quality index

The quality of the export basket is constructed following Haussman et al. (2007). First, they

propose an index called PRODY that attributes a level of productivity to each k (HS-6) line.

The total exports for a country i is,

�� =�����

���

And the level of productivity PRODYk associated to each k (HS-6 line) is constructed as

������ = ∑ ��� !�⁄ ∑ ��� !�⁄�

��� , (1B)

where Yi is the GDP per capita in Purchasing Power Parity of each country I,

∑ ���� ��⁄� isthesumoftheshareofproduct1exportedinallcountries.

This index is a variant of the Balassa’s index of revealed comparative advantage. This way,

exports from developed countries are considered as more productive that the ones coming from

developing economies.

Finally, the level of productivity associated to the export basket of each country i is,

���� = ∑ 5�� !� 6 ������� . (2B)

Thus, it depends on the degree of concentration of the export basket, weighted by the quality of

the products exported. The underlying idea behind this indicator is that diversifying its exports

basket away from products of low productivity may accelerate subsequent growth. We

compute a yearly EXPYi indicator.

The variety index

In order to allow comparability of the index between countries and time, the export variety (or

extensive margin of exports) is constructed following a modified version proposed by Feenstra

and Kee (2008) of the Hummels and Klenow (2005) index.

Hummels and Klenow (2005) propose a measure of “extensive margin” of trade that is

consistent with product variety for a CES function. This indicator can be defined as changes in

26

exports or imports that are due to changes in the number of goods (a change in the variety of

products) rather than changes in the amount purchased of each good. Besides the fact that this

formula is consistent with trade theory, we choose it among all the definitions of extensive

margin available in the literature review because it takes into account the importance of the

traded good instead of roughly counting lines.

The construction of the indicator is based on the idea that exports from countries h and F differ

but have some products varieties in common. This common set is denoted by

( ) ∅≠∩≡ Fit

hit JJJ . An inverse measure of export variety from country h will be defined by

∑

∑

∈

∈≡hitJj

hit

hit

Jj

hit

hit

hit jqjp

jqjp

J)()(

)()(

)(λ . (3B)

Therefore, the ratio

)(

)(

J

Jhit

Fit

λλ

measures the export variety of country h relative to country F. It

increases with the variety exported from country h, and decreases with the variety exported

from country F. Thus, to be measured, this indicator needs a consistent comparison country F.

Feenstra and Kee (2008) use the worldwide exports from all countries to the United States (US)

as benchmark. Indeed, US appear as the mayor partner in terms of imported variety (US

imports almost 99% of all the varieties existing) and provides highly disaggregated trade

databases (until 10 digit codes). Nevertheless, as Feenstra and Kee (2008) noted, it would be

preferable to use countries’ worldwide exports instead of US imports. Indeed, this restriction

makes the measure dependent to the import structure of the US. And for countries that export

goods that have a small value in the import structure of this partner or that do not export some

kind of varieties to it (mostly developing countries), the magnitude of their export variety will

appear under-evaluated. Thus, in order to correct for these effects we prefer to work with the

entire world as the benchmark F, as in Hummels and Klenow (2005), even if this forces us to

use only HS-6 desegregated trade data.

Moreover, we need a benchmark F that doesn’t change thought time, in order to associate any

variation in the indicator to a variation in the export variety of the country h. So, following

Feenstra and Kee (2008) we take the union of all products sold in the world market in any year

over the period 1980-2004, and we average real exports sales of each product over years. In

27

this way, hitth

Fi JJ ,∪≡ is the total set of varieties imported by the entire world in sector i over

all years, and )()( jqjp Fi

Fi is the average real value of world imports for product j (summed

over all source countries and averaged across years). Then, comparing country h to the world

(F) allows us to set 1)( =Jhitλ and the export variety by country h takes the form:

∑

∑

∈

∈=≡ΑFi

hit

Jj

Fi

Fi

Jj

Fit

Fit

hit

Fith

it jqjp

jqjp

J

J

)()(

)()(

)(

)(

λλ

. (4B)

Thus, export variety only changes due to variations in the numerator, and thus, due to changes

in the set of goods sold by the country h. This allows us to do comparison of export varieties

across countries and over time. Moreover, this indicator goes beyond a simple count of trade

lines, because it takes into account the relevance of the sector i (HS-6 line) in world trade.